USE OF ALTERNATIVE DATA:HIGHFREQUENCYREADOUTOF THE SITUATION - COVID POLICIES, MOBILITYAND R-NUMBER

Ashutosh Mani Dixit Suraj Regmi Economist Data Scientist [email protected] [email protected]

ABSTRACT The role of alternative data in the crisis was recognized even before the COVID-19 pandemic[1]. Now, the months of stalemate made it more urgent to understand the importance of high-frequency data to inform the policy responses [2]. In Nepal, the Government has exerted stay put measures, and physical data collection activities are suspended. The confirmed cases of COVID-19 has reached more than 560,000[3] and the country is on high alert . In this impasse, the number of secondary cases one would produce over the course of outbreak - the reproduction number (R0) is useful to monitor the transmissibility of COVID-19 [4]. As the R-value is rapidly changing, it can be affected by a range of factors, including not just how infectious a disease is but how Government responds to it, and how the population behaves1. The World Health Organization (WHO) has suggested to the Government of Nepal several recommendations to contain the further spread of COVID-19. To get a sense of how Nepal is coping with the coronavirus pandemic we look at the alternative data sets to get a better understanding of the pandemic policies, mobility, and R-value during COVID.

Keywords Alternative data · COVID-19 · R-Number

1 Objective

(a) To get the high frequency read out of the COVID situation in Nepal We calculate effective reproduction number (R-value) from OWID data (smoothed)[5], and gain additional insights from COVID-19 community mobility reports2, the Oxford Coronavirus Government response tracker - Oxford stringency index3 and Google search trends.[7] (b) Make available the source code for extracting alternative data The data, and source code, along with frequently updated dashboard monitoring the R-value will be made open and available for public use. arXiv:2109.00050v1 [stat.AP] 31 Aug 2021 2 Methodology

2.1 Effective Reproduction Number

The real time reproduction number is estimated using Bayesian approach, assuming the new number of daily cases satisfies the Poisson paradigm. The work[8] done by Kevin Systrom at the US state level is replicated here. Systrom used the modified version of a solution created by [9] to estimate a real time reproduction number. As with changing conditions (behavior of people, government policies, etc), the value of Rt changes. The effective reproduction number

1https://www.weforum.org/agenda/2020/05/covid-19-what-is-the-r-number/ 2Google community mobility report was launched in April to showcase change in mobility trends in COVID situations. 3Policy responses come from the Oxford Coronavirus Government Response Tracker (OxCGRT). The tracker is published by researchers at the Blavatnik School of Government at the University of Oxford[6] Alternative Data: COVID

depends on yesterday’s (or previous) reproduction number and number of daily new cases. [9] use Bayes’ rule to update the real time value of Rt from the number of daily new cases and prior value of reproduction number. The new number of cases are seen everyday. This number of new cases says us something about the tranmissibility. Also, the Rt value of today has relation with Rt−1 value of yesterday, and every previous value of Rt−m.

[9] use Bayes’ rule to update the true value of Rt based on the number of new cases daily. Mathematically,

P (k|R ) · P (R ) P (R |k) = t t t P (k)

So, if we see k new cases, the distribution of Rt is equal to the likelihood of seeing k new cases given Rt times the prior beliefs of the value of P (Rt) without the data divided by the probability of seeing this many cases in general.

Now, every day that passes, we use yesterday’s prior P (Rt−1) to estimate today’s prior P (Rt). The distribution of Rt is assumed to be a Gaussian centered around Rt−1, i.e. P (Rt|Rt−1) = N (Rt−1, σ), where σ is a hyperparameter.

Choosing a Likelihood Function P (k|Rt)

A likelihood function function says how likely we are to see k new cases, given a value of Rt. We model the probability of seeing k new cases according to Poisson distribution, with arrival rate λ equal to number of new cases each day.

λke−λ P (k|λ) = k!

Connecting λ and Rt

The connection between Rt and λ is given in the paper as:

γ(Rt−1) λ = kt−1e where γ is the reciprocal of the serial interval. The serial interval is about 7 days according to CDC. As we know the number of new cases on the previous day, we can reformulate the likelihood function as a Poisson parameterized by fixing k and varying Rt.

γ(Rt−1) λ = kt−1e

λke−λ P (k|R ) = t k!

3 Limitations

3.1 Google mobility data

There are blind spots in alternative data, in particular data coming from mobile phones. Lower smartphone penetration rate among older people, and rural population, may not give a complete picture of the mobility. Additionally, the Google mobility report states that their data comes only from android smartphone users who allowed the device to track their location.

3.2 Oxford Government Response tracker

Oxford Coronavirus Government response tracker does not aim to measure the appropriateness or effectiveness of a country’s response. So a higher index should not be interpreted as the efficacy or effectiveness of the policy.

2 Alternative Data: COVID

4 Background

After several months of relatively low COVID cases in Nepal , COVID-19 cases began to rapidly spike in mid-April 2021 following a steep upwards trajectory (Figure 1). Figure 1: Cases, stringency index4, and residential mobility5

Source: (Our World in Data); (Google LLC, 2020); (University of Oxford - Blavatnik School of Government, 2020)

4“The Stringency Index is an aggregate score/composite measure made up of a particular combination of policy indicators/response metrics from the codebook and their values (for the Stringency Index these are C1-8 and H1). The OxCGRT aggregates these policy indicator values into a common “Stringency Index” that runs from 0 -100.”- www.bsg.ox.ac.uk 5Stay at home requirements: 0 - no measures , 1 - recommend not leaving house , 2 - require not leaving house with exceptions for daily exercise, grocery shopping, and ’essential’ trips , 3 - require not leaving house with minimal exceptions (eg allowed to leave once a week, or only one person can leave at a time, etc) , Blank - no data

3 Alternative Data: COVID

It was on 29th April 2021 after the cases started to surge, the transmission rate shooted upto 2.480 on 21st April - highest observed by the country till date (Figure 2). Nepal reinstated the control of social behavior, the stringency index which measures the severity of Government response increased from 30.56 to 91.67, as the Government increased the stay home requirements from 0 to 3 i.e from “no measures” to “require not leaving house with minimal exceptions”. People spent more time at home, and there was a spike in “residential percentage change from the baseline6” from 1% on 28th April 2021 to 19% on 5th May 2021. Figure 2: R-value and stringency index

6All lines represent a 7-day moving average. Baseline values were established using a median of the corresponding day of the week from the period between January 3 and February 6, 2020.

4 Alternative Data: COVID

Grocery and pharmaceuticals - generally experiencing high mobility, also recorded slump in the movement on 5th May 2021 and thereon (Figure 3). On Wednesday 28th April 2021, the change in mobility in grocery and pharmaceuticals in Nepal was 81 percent, whereas on Wednesday 5th May 2021 it receded to -21 percent. This can also be interpreted as “people were 21 percent less likely to be in grocery and pharmaceuticals on Thursday 5th May” than they were in the baseline (median of the corresponding day between 3rd January and 6th February). Similarly, the people were 81 percent more likely to be in grocery and pharmaceuticals on 28th April 20217. This was a result of restriction on movement as the Government reduced the opening hours of Grocery. The index of restriction on internal movement went up to 2. Figure 3: Grocery and pharma percent change from baseline, and restriction on internal movement

Source: (Google LLC, 2020)[10]; (University of Oxford - Blavatnik School of Government, 2020)

7Because of the privacy, google did not release the absolute number.

5 Alternative Data: COVID

The movement in transit and stations changed from 56 percent in 28th April to -41 percent in 5th May from the baseline, approximately 43 percent points increase. The closure of public transport, and restrictions in internal movement imposed by the Government decreased the frequency of visits in the transit stations such as bus parks and airports. However, some movements were allowed with travel passes (Figure 4). Figure 4: Transit and station

Source: (Google LLC, 2020); (University of Oxford - Blavatnik School of Government, 2020)

6 Alternative Data: COVID

In the Google mobility report, the baseline for the weekend is the median of weekends falling between January 3 and February 6. As the weekend visits get closer to normal value the relative change becomes smaller, as such we see recurrent spikes in workplace mobility (Figure 5). In Nepal, while most of the corporate offices, IT companies, NGOs and INGOs, switched to work from home8, there were banks and financial institutions which continued to operate as per the regulatory instructions9 and remained open even during the lockdown with limited staff and reduced hours of operations. Figure 5: Workplace and mobility10

Source: (Google LLC, 2020); (University of Oxford - Blavatnik School of Government, 2020)

8https://kathmandupost.com/art-culture/2020/03/22/covid-19-and-the-shift-to-remote-working 9https://thehimalayantimes.com/business/banks-to-remain-open-in-kathmandu-valley-during-prohibitory-period 10Close public transport 0: No measures; 1: Recommended closing (or significantly reduce volume/route/ means of transport available) School closing 0: No measures; 1: Recommended closing; 2: Required closing (only some levels or categories, eg just high school, or just public schools; 3: require closing all levels.

7 Alternative Data: COVID

Furthermore, the schools closed and stopped in-person learning (Figure 5). Since most of the private school and students opted for distant learning11, the alternate data from Google trends reveal that the interest to download “zoom” and “google meet” - video communication software exceeded even the interest in “songs”, “music”, and “games” during the lockdown period (Figure 6). Figure 6: Google trends – interest over time12

Source: Google trends

5 R-value and control measures

The Government of Nepal has been quick to close the schools (Figure 7). The stay at home requirements, restrictions on gathering and cancellation of public events were made stringent only after the reproduction rate reached 2.48 on 21st April 2021. The scientific literature supports that the restriction on social behavior can work to break the chain of infection. It is accepted that more strict and timely restrictions have significant effects than slower, weaker ones. The radical control of social behavior has helped the Government of Nepal drop R down to about 0.450. But while it works on an average,

11https://www.nepalitimes.com/banner/lockdown-gives-distance-learning-a-boost-in-nepal/ 12Explore more here : https://trends.google.com/trends/explore?geo=NP&q=Zoom%20download,Download%20music,Download%20songs,Download%20games, Google%20meets

8 Alternative Data: COVID

there is no guarantee these measures will always work. Evidence from countries like Peru which suffered rising disease despite restrictive policies, reinforcing the fact that compliance and trust are also key to effectiveness13. Moreover, without vaccinating the majority of the population the reproduction number could shoot-up if people start losing patience with restrictions, or the Government eases the measures (Figure 7). As such there is an unprecedented challenge in front of scientists and policy makers in Nepal and around the world, the criteria for policy adjustments are unknown . Is cancelling the public events doing the heavy lifting, or closing schools? How much economic support would help pacify the situation? How far should the vaccination coverage be stretched ? There is an ample scope for future research to answer this policy dilemma to keep the pandemic control but at an acceptable economic and social cost14. Figure 7: R-value and Government Policies15

6 Conclusion

As the country is experiencing the data challenges during the COVID crisis, there is an urgent need to gear up efforts to use alternative data. The high frequency readout of the situation from alternative data can be useful for the Government to get the real time assessment of the R-value, and understand social behavior.

13https://theconversation.com/what-we-learned-from-tracking-every-covid-policy-in-the-world-157721 14https://www.nytimes.com/2020/04/06/opinion/coronavirus-end-social-distancing.html 15Vaccination policy: 0 - No availability , 1 - Availability for ONE of following: key workers/ clinically vulnerable groups (non elderly) / elderly groups , 2 - Availability for TWO of following: key workers/ clinically vulnerable groups (non elderly) / elderly groups , 3 - Availability for ALL of following: key workers/ clinically vulnerable groups (non elderly) / elderly groups , 4 - Availability for all three plus partial additional availability (select broad groups/ages) , 5 - Universal availability

9 Alternative Data: COVID

The Government of Nepal has been responding with the radical control of social behaviors and it has resulted in decreased mobility. Nepal’s COVID reproduction rate (R) is now down to 0.41, but if people start losing patience with restrictions or if the Government relaxes the control, R could quickly rise again. As such the vaccinating majority of the population becomes imperative. As the government tentatively eases lockdown restrictions around the world, it will be monitoring R very cautiously. The future research could be on finding the optimum policy response for the Government to tune their interventions quickly enough to stay ahead of the outbreak trajectory.

Acknowledgments

Authors would like to acknowledge the guidance received from Hiroki Uematsu, Senior Economist at the World Bank.

References

[1] Cornelia Hammer, Diane Kostroch, and Gabriel Quiros. Big data: Potential, challenges and statistical implications. Staff Discussion Notes, 17:1, 01 2017. [2] Louis Marc Ducharme, Tebrake James, and Zaijin Zhan. Keeping economic data flowing during covid-19, May 2020. [3] Ministry of Health and Population. [Online] Available at: https://covid19.mohp.gov.np/. [4] Jacco Wallinga and Peter Teunis. Different Epidemic Curves for Severe Acute Respiratory Syndrome Reveal Similar Impacts of Control Measures. American Journal of Epidemiology, 160(6):509–516, 09 2004. [5] Diana Beltekian Edouard Mathieu Joe Hasell Bobbie Macdonald Charlie Giattino Cameron Appel Lucas Rodés- Guirao Hannah Ritchie, Esteban Ortiz-Ospina and Max Roser. Coronavirus pandemic (covid-19). Our World in Data, 2020. https://ourworldindata.org/coronavirus. [6] Thomas Hale, Anna Petherick, Toby Phillips, and Samuel Webster. Variation in government responses to covid-19. Blavatnik school of government working paper, 31:2020–11, 2020. [7] Google 2021. Google trends. [Online] Available at: http://trends.google.com/.

[8] Kevin Systrom. Estimating COVID-19’s Rt in Real-Time, April 2020. [9] Luis MA Bettencourt and Ruy M Ribeiro. Real time bayesian estimation of the epidemic potential of emerging infectious diseases. PloS one, 3(5):e2185, 2008. [10] Google LLC "Google Community Mobility Reports". https://www.google.com/covid19/mobility/ Accessed: July 2021.

10